Rewrite rule breaks code when used with ghc -O

[relrod]

I have the following code:

module Main where

import Data.Thyme.Time

main :: IO ()
main = print $ (realToFrac (150 :: NominalDiffTime) :: NominalDiffTime)

• When I compile it with ghc --make -O0, I get 150s
• When I compile it with ghc --make -O, I get 0s

At first I thought it was a bug in GHC's optimizer, but then I saw the rewrite rules in Thyme and now I'm not sure where to file the bug, so I'm starting here. I'm able to reproduce this on GHC 8.0.1, 8.0.2, and 7.10.3.

Let me know if you need any more information, I'm happy to provide it or help test a fix.

[relrod]

BTW, if I use -ddump-simpl-stats, I see that the specific rule being fired is realToFrac/NominalDiffTime-Fractional

[relrod]

@liyang This creates a difference at runtime based on the level of optimization used with GHC - it's not caught during typechecking or anything, so I'm bumping this...

(I worked around it by figuring out how to get rid of the realToFrac call in my code entirely, but given how subtle the bug is, it might be good to get a fix out).

[relrod]

Bumping this again...

[relrod]

And again.

[relrod]

Yearly bump, but guessing this package just isn't maintained at this point.

[relrod]

Hm, missed last year. Yearly bump 2x.

[ysangkok]

The package is now collectively maintained.

I wonder if this is the cause of the test failures in

• Compatibility with GHC 9.8 and 9.10 #77

It seems that cabal-install is passing -O by default.

[jamesdbrock]

Hi @relrod I sent you an ownership invitation for the Github Haskell Trust. If you accept then you can merge PRs and publish any package in the Trust, including this one.

Thank you for your attention to this package @ysangkok .